We review Bayesian optimization as a data-driven tool to select hyperparameters in econometric models. In particular, we consider the steady-state Bayesian vector autoregression, Villani (2009), where the hyperparameters guard against overfitting. The choice has to balance well between "over-shrinking" and overfitting to yield reliable inference and predictions for the researcher in different settings. The common approach has been to rely on standard values that have been used no matter the application. However, there has been a renewed interest in finding the optimal hyperparameters for a given problem, e.g. Giannone et al. (2015), and Carriero et al. (2012).

Despite good results with machine learning applications for over a decade (e.g. Brochu et al., 2010) and Snoek et al., 2012), Bayesian optimization has rarely been used for optimization in econometric models. Also, we propose a new acquisition strategy that exploits the duration-precision trade-off within MCMC.

The idea is illustrated by finding optimal values for the steady-state-BVAR model using seven US macroeconomic time series.

References:
Brochu, E., Cora, V. M., and De Freitas, N. (2010). A tutorial on Bayesian optimization of expensive cost functions, with application to active user modeling and hierarchical reinforcement learning. arXiv preprint arXiv:1012.2599.

Carriero, A., Kapetanios, G., and Marcellino, M. (2012). Forecasting government bond yields with large Bayesian vector autoregressions. Journal of Banking & Finance, 36(7):2026_2047.

Giannone, D., Lenza, M., and Primiceri, G. E. (2015). Prior selection for vector autoregressions. Review of Economics and Statistics, 97(2):436_451.

Snoek, J., Larochelle, H., and Adams, R. P. (2012). Practical Bayesian optimization of machine learning algorithms. In Advances in neural information processing systems, pages 2951_2959.

Villani, M. (2009). Steady-state priors for vector autoregressions. Journal of Applied Econometrics, 24(4):630_650.

 

Tid: 30 oktober, 2019, kl. 13-14 Plats: B705

Välkommen.